A flake8 plugin to check Pydantic related code.
MIT License
A flake8
plugin to check Pydantic related code.
flake8_pydantic
parses the AST to emit linting errors. As such,
it cannot accurately determine if a class is defined as a Pydantic model. However, it tries its best, using the following heuristics:
BaseModel
or RootModel
.model_config
attribute set.Field
function.Annotated
.computed_field
or model_validator
.model_dump
.PYD001
- Positional argument for Field default argument
Raise an error if the default
argument of the Field
function is positional.
class Model(BaseModel):
foo: int = Field(1)
Although allowed at runtime by Pydantic, it does not comply with the typing specification (PEP 681) and type checkers will not be able to synthesize a correct __init__
method.
Instead, consider using an explicit keyword argument:
class Model(BaseModel):
foo: int = Field(default=1)
PYD002
- Non-annotated attribute inside Pydantic model
Raise an error if a non-annotated attribute is defined inside a Pydantic model class.
class Model(BaseModel):
foo = 1 # Will error at runtime
PYD003
- Unecessary Field call to specify a default value
Raise an error if the Field
function
is used only to specify a default value.
class Model(BaseModel):
foo: int = Field(default=1)
Instead, consider specifying the default value directly:
class Model(BaseModel):
foo: int = 1
PYD004
- Default argument specified in annotated
Raise an error if the default
argument of the Field
function is used together with Annotated
.
class Model(BaseModel):
foo: Annotated[int, Field(default=1, description="desc")]
To make type checkers aware of the default value, consider specifying the default value directly:
class Model(BaseModel):
foo: Annotated[int, Field(description="desc")] = 1
PYD005
- Field name overrides annotation
Raise an error if the field name clashes with the annotation.
from datetime import date
class Model(BaseModel):
date: date | None = None
Because of how Python evaluates annotated assignments, unexpected issues can happen when using an annotation name that clashes with a field name. Pydantic will try its best to warn you about such issues, but can fail in complex scenarios (and the issue may even be silently ignored).
Instead, consider, using an alias or referencing your type under a different name:
from datetime import date
date_ = date
class Model(BaseModel):
date_aliased: date | None = Field(default=None, alias="date")
# or
date: date_ | None = None
PYD010
- Usage of __pydantic_config__
Raise an error if a Pydantic configuration is set with __pydantic_config__
.
class Model(TypedDict):
__pydantic_config__ = {} # Type checkers will emit an error
Although allowed at runtime by Python, type checkers will emit an error as it is not allowed to assign values when defining a TypedDict
.
Instead, consider using the with_config
decorator:
@with_config({"str_to_lower": True})
class Model(TypedDict):
pass
And many more to come.
Once the rules of the plugin gets stable, the goal will be to implement them in Ruff, with autofixes when possible.